AI-generated tests in a regulated environment face unique challenges, and Fresenius’ approach, with its PoC named Intricor, uses AI as an assistant to support testers.

Richard Seidl Software Testing

This is the other podcast on Software Testing by Richard Seidl, the episodes are in spoken German but the show notes and site are written in English. Our summaries are generated from AI transcript translations.

• https://www.richard-seidl.com/en/blog/tag/podcast-software-testing
• https://www.richard-seidl.com/en/

Episode Details

• Show Notes: https://www.richard-seidl.com/de/blog/ki-generierte-testfaelle
• Published: 2026-02-24T05:00:00Z
• Duration: 1430
• Author: Richard Seidl - Experte fur Software-Entwicklung und Testautomatisierung

Overview

The podcast explores how AI is being integrated into test automation in the highly regulated medical device industry, with a focus on Fresenius Medical Care’s development of AI-assisted test generation for hemodialysis machines. A key component of this effort is the Proof of Concept system called Intricor, which uses AI to support human testers in creating test scenarios rather than replacing them. The system is built around principles such as traceability, structured data management, and human oversight to ensure compliance with stringent regulatory requirements.

Challenges faced in this integration include managing physical system dependencies, ensuring the accuracy of AI-generated test cases, and maintaining compliance through required manual validation processes. The approach taken emphasizes adaptability, modularity, and the importance of incorporating expert input to refine AI outputs. The discussion also suggests potential future developments, such as expanding the system to include more hardware components and integrating it with product lifecycle management systems to enhance its utility and compliance in the medical device field.

What If

• What if you develop a lightweight AI test assistant with traceability for regulated domains?

  • Concrete Move: Build a “one-click generator” that uses LLMs to auto-generate test cases for regulated software, with built-in traceability to documentation.
  • Why Now: Legacy systems and outdated ALM tools create resource bottlenecks, and AI can reduce test case generation time from days to minutes.
  • Expected Upside: Accelerate testing cycles while meeting regulatory compliance, similar to Freseniuss Intricor PoC, enabling faster time-to-market for critical systems.

• What if you implement a human-in-the-loop validation workflow for AI-generated test scenarios?

  • Concrete Move: Design a system where AI-generated tests are reviewed by domain experts, with formal e-signature approval and logging of changes in a REC system.
  • Why Now: Regulated industries require traceability and avoid AI hallucinations, and manual validation ensures accuracy without compromising automation.
  • Expected Upside: Maintain compliance while leveraging AI efficiency, reducing rework and increasing stakeholder trust in AI outputs.

• What if you create a modular architecture for AI testing that integrates with PLM/ILM systems?

  • Concrete Move: Develop a flexible AI test framework that connects to PLM/ILM via APIs, enabling keyword-driven test scripts and real-time data synchronization.
  • Why Now: Fresenius highlights the need for ILM/PLM integration, and modular systems allow rapid model updates (e.g., ChetGPT 5, Lama) without overhauling the entire architecture.
  • Expected Upside: Streamline testing across software and hardware-in-the-loop systems, reducing complexity and enabling future scalability in regulated environments.

Takeaway

• Implement AI-assisted test generation with human validation: Use AI tools like Intricor to create test scenarios, but pair them with a human-in-the-loop process for review and e-signature approval, ensuring compliance in regulated environments.
• Build a modular AI architecture with adaptability: Prioritize modularity in your AI system to easily swap models (e.g., LLMs) as needed, avoiding reliance on a single model that may become obsolete.
• Automate prompt generation for AI testing: Develop a one-click test case generator using LLMs to optimize prompts and minimize user input, reducing manual effort in test creation.
• Ensure traceability in AI-generated tests: Centralize test data using a system like the “wreck system” and document every process step with traceability, linking test expectations to referenced documents for audit readiness.
• Integrate AI testing with PLM/ILM systems: Connect your AI test tools to existing document management and product lifecycle systems via service interfaces, improving accessibility and aligning with regulatory documentation needs.